Open letter: our public comment on Colorado's AI rules

Submitted to the Colorado Department of Law on October 22, 2026, ahead of the October 26, 2026 rulemaking hearing. Published here as an open letter so anyone can read it.

To: Colorado Department of Law, Consumer Protection Section
From: Rowe Heikkinen, Founder, Constitution for Human Creation
Re: Proposed Rules 4 CCR 904-6, Automated Decision-Making Technology & Conversational Artificial Intelligence Service Rules, implementing the Automated Decision-Making Technology Act (ADMT Act, SB26-189) and the Chatbot Safety Act (HB26-1263). Comments address the August 11, 2026 proposed rules and the October 6, 2026 interim draft.

On this page: The letter in plain words · The full comment as filed

The letter in plain words

This is a short, plain-language version of each part. The full comment we filed follows below.

Who we are

Constitution for Human Creation is a volunteer-led Colorado civic project inviting everyday people to help shape a public charter for how AI should be governed. It's a living draft, not law, and no one has endorsed it. We intend to become a formal organization, with leadership chosen from among our contributors by an open public vote. The rules already do much of what our principles ask for, and our five suggestions are meant to strengthen them.

1. A named, accountable person

Principle: Named Human Responsibility.

Companies using AI should stay responsible even when someone else runs the system. We ask the state to bring back the standard that, whenever feasible, the person reviewing a decision isn't the one who made it or that person's subordinate, and to have each company name an accountable person or office for each AI system.

2. Records made at the time

Principle: Keep Systems Auditable.

When a company claims its AI didn't really affect a decision, records made at the time of the decision should count most. Explanations written afterward shouldn't be enough on their own.

3. Chatbot honesty you can check

Principle: No Systematic Deception.

Chatbots have to say they're AI. We ask that companies' annual reports summarize how they tested whether people actually understood that, what they found, and what they changed.

4. Real recourse before harm

Principle: Digital Does Not Override Physical.

When an AI decision would deny someone food, shelter, or needed medical care and would be hard to undo, the company should pause it where feasible while a correction or review is pending, or write down why it couldn't.

5. The people who do the work

Principle: Professionals and the public interpret the principles together.

Chatbot companies should say whether licensed mental-health professionals reviewed their self-harm protocols. As the rules develop, the state should hear from nurses, teachers, housing counselors, HR staff, and lending officers, not only developers and lawyers.

The full comment as filed

Thank you for the chance to comment, and for the clear, example-driven drafts your office has published.

My name is Rowe Heikkinen. I founded Constitution for Human Creation, a volunteer-led Colorado civic project that invites everyday people to help shape a public charter for how AI should be governed. It pairs eight short Foundational Principles for AI with a longer living charter that grows through public feedback. It is a living draft. It is not law, it does not replace applicable law, and no organization or person has endorsed it. We are not a nonprofit or an industry group. We intend to become a formal organization, with leadership chosen from among our contributors by an open public vote.* I am writing as a citizen, and nothing here is legal advice.

One set of rules implements two laws, so each rule we cite below is labeled with the act it implements. Rule numbers follow the October 6 interim draft unless noted as August numbering. Overall, these rules already reflect much of what our principles ask for: plain, accurate communications (Rule 3.2, which applies to both acts), specific principal reasons in adverse outcome notices (Rule 7, ADMT Act, SB26-189), meaningful human review (Rule 8.7, ADMT Act, SB26-189), and chatbot disclosure and annual reporting (Rules 11 and 14, Chatbot Safety Act, HB26-1263). The five suggestions below are meant to strengthen what is already there.

1. Keep a named, accountable person at the center of human review.

Principle: Named Human Responsibility. “Every serious deployment has a named human owner. Liability follows authority. ‘The system did it’ is never a sufficient answer.”

We support Rule 4.3(A) (ADMT Act, SB26-189), which makes the Deployer responsible for all Deployer obligations when a third party runs the Covered ADMT, and Rule 8.7(C)(1)(a)(1) (ADMT Act, SB26-189), which says the Deployer “remains fully accountable” when it uses an outside reviewer. We are concerned that the interim draft dropped the August draft’s standard (then Rule 7.7(C)(1), ADMT Act, SB26-189) that the reviewer be someone who did not make the original decision and is not that person’s subordinate, whenever feasible.

Suggestion: Restore that independence standard “whenever feasible,” alongside the new internal-or-external option. Also ask each Deployer to designate a named accountable person or office for each Covered ADMT, recorded with the Rule 8.7(H) (ADMT Act, SB26-189) documentation and available to the Department on request.

2. Rebut the material-influence presumption with records made at the time.

Principle: Keep Systems Auditable. “Harm must be traceable to causes and people so it can be understood, stopped, and prevented from repeating.”

Your October 6 cover page asks what alternative to the rebuttable presumption would work, “outside of the Deployer simply asserting” that an output was de minimis. We support keeping the presumption in Rule 6.1 (ADMT Act, SB26-189). Rule 6.1(B)(5) (ADMT Act, SB26-189) already mentions “contemporaneous evidence,” and Rule 8.7(H)(1) (ADMT Act, SB26-189) now requires “contemporaneous documentation” of human reviews.

Suggestion: State that evidence created at or before the time of the decision (for example, a decision-maker’s recorded independent judgment, or a published, consistently applied cut-off) carries the most weight, and that after-the-fact statements alone generally will not rebut the presumption. We also support keeping the consideration in the Rule 2 definition of “De Minimis Factor” (ADMT Act, SB26-189) about whether a Deployer’s marketing, contracts, internal policies, or training materials direct staff to treat the output as primary or authoritative.

3. Make chatbot honesty measurable.

Principle: No Systematic Deception. “Do not build or run systems that systematically deceive or manipulate people.”

We support Rule 3.2(A)(7) (applies to both acts; communications must not be unfair, deceptive, false, or misleading), the AI disclosure requirements in Rule 11.2 (Chatbot Safety Act, HB26-1263), and Rule 13.2 (Chatbot Safety Act, HB26-1263) on outputs or profiles that imply a licensed professional. For audio disclaimers, Rule 11.2 (Chatbot Safety Act, HB26-1263) already lets the Department consider whether the Operator tested users’ understanding that they are interacting with AI and whether changes were made after testing.

Suggestion: Since the Chatbot Safety Act lets the Attorney General determine additional metrics on the efficacy and reliability of implemented safeguards for the annual report, add to Rule 14 (Chatbot Safety Act, HB26-1263) a short summary of the Operator’s user- awareness testing for its AI disclosures: what was tested, the results, and what changed. That turns a good disclosure rule into something the public can check over time.

4. Real recourse before harm becomes irreversible.

Principle: Digital Does Not Override Physical. The digital layer “does not override physical reality or people’s lives without clear consent and real recourse.”

The interim draft removed the August draft’s provisions that, where possible, an Adverse Outcome be stayed pending correction of data or pending human review (then Rules 7.4(D) and 7.7(G)(1), both ADMT Act, SB26-189), and the presumption that review is commercially reasonable for a “severe and irreversible denial of a basic human need” (then Rule 7.7(E)(3), ADMT Act, SB26-189). We appreciate that the interim draft keeps the Deployer’s burden of proof (Rule 8.7(E)(3), ADMT Act, SB26-189), names food, shelter, and required medical care in the harm factor (Rule 8.7(E)(2)(b), ADMT Act, SB26-189), and requires an explanation when review is declined (Rule 8.7(F)(1)(e), ADMT Act, SB26-189).

Suggestion: Restore a narrow version of that protection. When an Adverse Outcome would deny food, shelter, or required medical care and would be hard to undo, the Deployer should pause the outcome where feasible while a timely correction or review request is pending, or document why it could not.

5. Let the people who do the work help shape the rules for their field.

Principle: Our project holds that principles should be interpreted by “a hybrid of professionals who understand the systems and public representatives who live with the consequences.” We don’t think a founder, or any one group, should decide how AI is used in medicine, housing, or the classroom.

The rules already lean on field expertise: the “Evidence-Based Method” definition in Rule 2 (Chatbot Safety Act, HB26-1263) names established clinical screening tools, Rule 14.3(A)(3) (Chatbot Safety Act, HB26-1263) asks how self-harm protocols incorporate “clinical best practices and research,” and Rule 8.7(C)(2)–(3) (ADMT Act, SB26-189) requires reviewers with subject-matter understanding and training.

Suggestion: (a) In Rule 14.3(A)(3) (Chatbot Safety Act, HB26-1263), ask Operators to state whether licensed mental-health or suicide-prevention professionals reviewed their protocols. (b) As the Department refines the sector examples (Rules 6.2 and 7.5, both ADMT Act, SB26-189) and any later guidance, invite working practitioners such as nurses, teachers, housing counselors, HR staff, and lending officers, in addition to developers and counsel.

Thank you for considering these comments. We would be glad to share more input, including feedback from the professionals and residents who take part in our open Monday meetings. You can reach me at Rowe.heikkinen@chcag.org. The principles are published at chcag.org.

* This line was clarified after filing. The comment as filed said "leadership chosen by open vote of our contributors." What we mean is that the public votes on the people who contribute.

Respectfully,
Rowe Heikkinen
Founder, Constitution for Human Creation
chcag.org · Rowe.heikkinen@chcag.org